Polaris Refresh (RX 500 Series) Rumors

[lobz]

And 9Ghz cant cost that much more if nv using them on low end GTX1060(yes gtx1060 is low end now)

great, you've just successfully ended all useful conversation here for a while with this simple sentence...

 

[mohit9206]

I hate rebrands. Amd is especially terrible at it. You have 7790 > R7 260X > R7 360..
Then you have 7850 > R7 265 > R7 370.
Rebranding once is understandable but amd loves to rebrand over and over again as long as it can keep fooling people.

Tough to be excited about a 10-15% performance increase a year later. These cards should be priced less than 480/470's. $300 for a highly overclocked polaris card? 1070's are down to $320 already in a different class of performance.

Amd doesn't have a magic wand that they can give 30-40% increase from the same architecture and same specifications. 10-15% is fine as long as the generational price go down. But knowing the track record of amd, rx580 will be $240 while rx480 will be available for less than 200, rx570 will be $180 while rx470 go for less than 150 etc.
Hopefully the rx560 is a big improvement over rx460 with more shaders but even there am not expecting more than 20% increase.

looks like I'll just hang on to the 480 until something else arrives

Good to know you're still hanging onto the gtx 480.

 

[guskline]

Though I have 2 original reference RX480-8g gpus that I paid @$239 each for last year when they came out, I have no problem with AMD rebranding to the 500 series. The baseclocks are up and hopefully the latest improvements have been refined for this chip.

Waiting for Vega to go public.

 

[Mopetar]

Someone on reddit suggested the base clock was 1340, and the boost clock was 1425. Do you guys think it's actually possible that AMD squeezed out 12.5% more clock speed out of Polaris?

I'll believe that the GloFo process was bad enough that a 12.5% performance bump solely off the back of improvements there is
certainly possible.

 

[Bacon1]

I'll believe that the GloFo process was bad enough that a 12.5% performance bump solely off the back of improvements there is
certainly possible.

You can see people hitting high on 480s now:

https://www.reddit.com/r/Amd/comments/63louu/gpuz_rx_580_from_chiphell_leaked/dfvintx/

Anyway I don't see this series being that important

 

[Mopetar]

You can see people hitting high on 480s now:

https://www.reddit.com/r/Amd/comments/63louu/gpuz_rx_580_from_chiphell_leaked/dfvintx/

Is it any surprise that they're hitting the kind of clock speeds that AMD will soon sell in a 580 well after the launch? The post you linked even says 2017, which suggests its a newer chip than the launch chips.

If there's no difference between a 580 and 480 based on fab process, it just comes down to binning. The 580 is basically what would have been sold as a *80 X version (although not quite the same as the X versions typically had more SPs and not just a clock speed advantage) in previous generations. I don't think the 480 has any fused off parts so its just clock improvements.

 

[Bacon1]

Is it any surprise that they're hitting the kind of clock speeds that AMD will soon sell in a 580 well after the launch? The post you linked even says 2017, which suggests its a newer chip than the launch chips.

If there's no difference between a 580 and 480 based on fab process, it just comes down to binning. The 580 is basically what would have been sold as a *80 X version (although not quite the same as the X versions typically had more SPs and not just a clock speed advantage) in previous generations. I don't think the 480 has any fused off parts so its just clock improvements.

Yeah the major change will probably be for reviewers - no throttling reference card. That hurts them the most in reviews when it can't even hold 1266 and hovers around 1100s. Since people are getting 1300-1400 clocks from custom 480s right now thats a lot of performance they are missing in reviews. You'd think they'd have learned from the poor 290(x) reference cards . 390(x) reviewed much better because they were all custom cards.

 

[Mopetar]

I can see the point of putting out a reference card, but it's probably not a bad idea to launch along side third party cards and ship at least one of those to every reviewer as well. It probably means that more will leak out ahead of time, but that seems to happen no matter what you do these days.

I think it would be really interesting if AMD/NV sold the rights to the reference card to some third party. Everyone still gets to put out their own custom cards, but there's at least better odds of the reference card being better. If they're worried about showing favoritism, just let there be different companies doing the reference card for each different GPU.

 

[w3rd]

Or is Dr Su to be sandbagging...? and dropping "leaks" into channels, sniffing for liable leaktardz..

Aren't these rumors all so fascinating..?

 

[unseenmorbidity]

1450 boost clock. I guess I shouldn't be that impressed since glofo made the first batch.

 

[Head1985]

It will be memory bandwidth bottleneck...

 

[Head1985]

LOL so now low end range is priced at +250$/+300€? Good old Nvidia fooling people...

Yes GTX1060 was mid range year ago when launched.Today after GTX1080TI launch and GTX1080 price cut its just low end performance wise.

GTX1080TI/TITANXP
GTX1080
GTX1070
GTX1060-low end
GTX1050 entry

 

[w3rd]

Nv dont need do anything.They already have GTX1060 out and thats enough for rx580...

Btw 1500mhz is good clock.Its finaly at maxwell level and it should have same performance as GTX980 1500Mhz.Maxwell GTX980/970 is still faster after oc to 1500mhz than max oc rx 480/470 so rx 570/580 should close the gap.Lilttle late tho...
On other hand pascal have zero oc headroom and GTX1060 max oc is slower than GTX980 1500Mhz so rx580 might be faster max oc vs max oc vs GTX1060 atleast.

Still i dont get why they didnt add 9Ghz memory..After oc to 10Ghz rx580 at 1500/10000 would be beast.AMD cant do anything perfect.Always there is huge mistake.

I did preface my comments about gamers and DX & Vulkan.

Coincidentally, the GTX1060 doesn't do well in Windows 10's dX12 games, nor in Vulkan. The 1060 also doesn't have AMD's driver package, or even CGN. The GTX1060 is also missing freesync.

All of which I prefer. (even though I rock a sweet NVidia rig)

You have to be soft in your head to say that RX 580 will compete with GTX 1070. The difference in performance is too big for DX11 games to higher core clock on RX 580 mitigate the difference. In Doom(Vulkan) it can be close, but still will not be better option than GTX 1070.

Read above. I did mention the games I was most interested in. And even the best engine will support such features, so I am interested in unified support. AMD tech is in xbox & playstation. And the xbox scorpio this xmass will be the blow out seller of the year.

I expect games companies to adopt AMD tech rapidly.

We don't know that for sure though as the 580 has an 8-pin power connector or an additional 75W to play around with. It's also possible that the 580 uses slightly more power than the 480 to achieve its baseline 1340 MHz, but just has substantially more room for an overclock. It's also possible that the 580 is able to hit that 1340 MHz for less power than the 480 needed for its baseline, but still doesn't have a whole lot of OC headroom and is just using the 8-pin connector so it doesn't have the same launch fiasco as the 480 which could overdraw from the PCI-E bus.

Betting on a 580 that uses less power for better baseline clock speeds while having substantially more overclock headroom is choosing the best case scenario at this point. It's certainly possible that is exactly what we get, but believing in it so heavily is really just setting yourself up for disappointment later when it just turns out that its a mix of both bad and good.

You just spent a whole post narrating the worst case scenario, then saying: "Betting on a 580 that uses less power for better baseline clock speeds while having substantially more overclock headroom is choosing the best case scenario at this point..."

Oddly, I would say that^ is the best conclusion one can make. Given all available knowledge, given AMD's own remarks, leaked slides and forthcomings. They have been mum on the specs, but not their purpose. You seem lost bro... I am so glad we can help.


Additionally, the best case scenario would be, if the Radeon RX580 stays the same discounted prices as the outgoing Radeon 480's...!

And sits in a new slow-spun wafers.

 

[Ansau]

Yes GTX1060 was mid range year ago when launched.Today after GTX1080TI launch and GTX1080 price cut its just low end performance wise.

GTX1080TI/TITANXP
GTX1080
GTX1070
GTX1060-low end
GTX1050 entry

Yeah, you confirm that you are fooled by Nvidia chip branding when deciding gpu tiers, like most of people out there and media.

Nvidia's chip branding (GP104, GP106...) is just that, chip branding. Nvidia has never had a trend about having each chip targeted to specific gpu tiers. Here are some examples:
- Tesla had 2 chip generations, G9X and GT2XX, the second introduced in GeForce 200 series.
- In Tesla, Nvidia used the biggest chip (GT200) for 5 gpus (from GTX260 to GTX295).
- In Fermi, sub 200$ gpus were still using the second biggest chip, GTX 460 SE (160$) and GTX 560 (199$).
- In Kepler, Nvidia changed how they were releasing gpus based on the chips. Before, gpus with the biggest chip were among the first to be released, but since Kepler biggest chip has been released quite after smaller chips are introduced. GK104 kept being the biggest Nvidia chip produced for over a year.
- Kepler supposed a premium price tag for gpus with the biggest chip, going from 500$ to 1000$ for the most expensive consumer gpu.
- Maxwell supposed a heavy reduction on the amount of gpus created from all chips. Maxwell also had 2 chip generations, but in this case only the smallest one (GM107) was included in the first, and the rest in the second.
- For the first time, in Pascal the biggest chip (GP100) is only available to the professional market. Consumers have, so far, access to a smaller brother to the biggest chip (GP102) with reduced amount of CUDA cores (3584 vs 3840 of full GP100 and GP102), cutted FP64 performance and without other new technologies (GDDR5X instead of HBM2, lack of NVLink).
- GTX x60 have been based on both Gx104 and Gx106 chips, depending if these chips have been more or less powerful in each generation.

It is clear saying a gpu is in a tier or another based on which chip uses is clueless. Nvidia has never used chip branding for tiering gpus, it has always been the media and public trying to make a rule, because it seems people cannot live without dumb segmentation.

GTX 1060 is no way a low end gpu. Neither its price nor its performance can be considered low end tier, characteristic for low priced gpus used in tight budgets. You differentiate low end and entry level, but they are essentially the same, cheap gpus used in cheap systems and for an audience that can live without ultra settings at +60fps.

You are all fooled because Nvidia used to use the second biggest chip for GTX x60, but Maxwell and Pascal use its smaller brother. But that doesn't put the 960 or the 1060 into the low end tier. They are both mainstream/mid tier (960 when it was launched). Gpus with reasonable prices that have good enough performance to play most of the games at ultra or those more demanding with minor setting tweaks, and to still be comfortable in very high resolutions with lesser demanding games.

Pascal line is more like this:
GTX1080TI/TITANXP - Enthusiast/premium
GTX1080/1070 - High end
GTX1060 - Mid end/mainstream
GTX1050/ti - Low end/entry level

 

[Head1985]

1-GP100/GP102 have same number of SP.GP102 is just pure gaming version of GP100 without fp64.
2-Midrange is not midrange after one year and with new products like 1080Ti out.All nv gpus after 1080Ti launch are one tier bellow launch.Its same with amd rx480(232mm2) is only 7870(212mm2) successor they didnt release 7970 or 290x successor yet.

 

[nathanddrews]

NVIDIA markets the GT730 and GT710 as entry-level DX12 cards, not the 1050.

 

[CatMerc]

1450 boost clock. I guess I shouldn't be that impressed since glofo made the first batch.

Fake

 

[Mopetar]

You just spent a whole post narrating the worst case scenario, then saying: "Betting on a 580 that uses less power for better baseline clock speeds while having substantially more overclock headroom is choosing the best case scenario at this point..."

If you're explaining why something is an unlikely best case scenario, it helps to contrast it with the different possibilities, which include using more power and not having substantial OC headroom beyond 1500 MHz. But just like the best case outcome, the worst-case isn't all that likely either. It's more likely to fall somewhere in the middle, especially as the number of factors increase.

You especially seem to view the world though some rose colored lens where AMD will always have things go right for them. If you were posting here prior to the Polaris launch, you'd probably have been one of the people claiming that it was going to be amazing and have Fury-like performance. The 580 is more likely than not to turn out to be just an average upgrade, which is really okay considering the 480 was a bit under the curve, at least in terms of the months around launch. Sure, it could turn out to be really amazing, but without evidence or good reason to believe that case to be likely, it's just wishful thinking.

 

[w3rd]

It doesn't have to be any of that hocus.
Everything you said... is just how you think it's going to be, based on knowing nothing of what Dr Su has said, what AMD engineers have suggested, and then pretending not to know, or see recent leaked slides. Your posts are essentially ignorant of current knowledge. Then in a trolling fashion, you use your new found ignorance to spew FUD. So later you can come back and say... "I didn't know..?"

Seems, you have no evidence of anything, because you are not looking for it..

Understand, Your downtrodden predictions are quite comical, as you are PURPOSELY voicing a worse case scenario in everything AMD, and your excuse is because later on, You don't want "others" to be disappointed..?


When in reality, You... just don't want to be disappointed.

Insulting other members is not allowed.
Markfw
Anandtech Moderator

 

[PhonakV30]

https://i.redd.it/zwdjeed0rqpy.jpg

1450 boost clock. I guess I shouldn't be that impressed since glofo made the first batch.

Win 7 and while SS before installing 17.4.1 (XFX 290 Double dissipation)

3 things :
1) Polaris 20
2) AMD Crossfire
3) Revision E7 (!!)

 

[Bacon1]

Everything you said... is just how you think it's going to be, based on knowing nothing of what Dr Su has said, what AMD engineers have suggested, and then pretending not to know, or see recent leaked slides. Your posts are essentially ignorant of current knowledge. Then in a trolling fashion, you use your new found ignorance to spew FUD. So later you can come back and say... "I didn't know..?"

I don't think they've said anything about Polaris refresh?

 

[w3rd]

There have been several mentions when talking about Vega, or an engineer mentioned that further work on Polaris, will exhibit lessons learned when working on Vega...

What then is Polaris 20?

 

[EXCellR8]

might actually try to see if I can score a pro duo as an upgrade to the RX 480. I know it's older but since CF doesn't work for sh1t it seems the best route for a noticeable increase in performance--adhering to team AMD anyway

 

[Mopetar]

Everything you said is just how you think it's going to be, based on knowing nothing of what Dr Su has said, what AMD engineers have suggested, and then pretending not to know, or see recent leaked slides.

I'm not going to trust a company to give it to me straight. If they wanted to dispel all notion of whether or not they were being truthful, they'd demonstrate a benchmark. Otherwise, I'm going to assume that they want to sell me a product. There's nothing wrong with that. Apple CEO and engineers talk about how great their product is, Google CEO and engineers talk about how great their product is, Microsoft CEO and engineers talk about how great their product is. No one talks about the awful turd they're making, and really only acknowledge such if it's so bad that no amount of company Kool-aid can bend reality.

Your posts are essentially ignorant of current knowledge. Then in a trolling fashion, you use your new found ignorance to spew FUD. So later you can come back and say... "I didn't know..?"

Spare me, please. Provide links to this "current knowledge" you have. If you think anyone behaving rationally and thinking critically is trolling this place must seem like a very large bridge to you.

Seems, you have no evidence of anything, because you are not looking for it..

Seriously, provide links to such evidence. All of the linked rumors so far suggest that my assumptions are reasonable.

Understand, Your downtrodden predictions are quite comical, as you are PURPOSELY voicing a worse case scenario in everything AMD, and your excuse is because later on, You don't want "others" to be disappointed

Please find where I've said it's going to be nothing but doom and gloom or the worst case. Pointing out that such a case exists, doesn't mean I believe it to be likely. As I said before, it's about as unlikely as a best case scenario, but you seem to read what you want to read and ignore anything else. Then you accuse me and others of making comical predictions when your own posts seem to claim that the 580 will have amazing performance, good OC headroom, at low power, while AMD sells them at a discounted price. I'd swear that this has to be deliberate as I can't imagine how someone could honestly think this way.

 

[frozentundra123456]

Yes GTX1060 was mid range year ago when launched.Today after GTX1080TI launch and GTX1080 price cut its just low end performance wise.

GTX1080TI/TITANXP
GTX1080
GTX1070
GTX1060-low end
GTX1050 entry

I would base category on "performance" not price. And in that whole year there was only one card (except for the very niche Titan) that offered higher performance than the market had when the 1060/1080 came out, and that of course was the 1080Ti. Even then the average performance gain was only 25% or so, hardly enough to redefine the market. I would consider 1080/1080Ti high end, 1060/470/480 midrange, and anything below that low end/entry level. The 1070 is kind of a "tweener", that falls between the midrange and high end cards.